Undo doesn't undo from the first initial change to unmodified file

[jpoirier]

• VSCode Version: Code 1.15.0 (8b95971, 2017-08-09T20:09:15.006Z)
• OS Version: Linux x64 4.12.5-300.fc26.x86_64
• Extensions:

Steps to Reproduce:

1. Add a single modification to an unmodified file, e.g. add a newline

2. Undo the modification via Ctrl+Z

3. The file stays tagged as modified but should go back to its unmodified state without having to save it; doing a save means a formatter may reformat the file, which is not what's desired.

   E.g. I open a project's json file and add some info to it then hit save but the file gets reformatted when I save it, the new formatting is undesired so I click undo to get back to the file's original state. The file goes back to the original state text wise but is still tagged as modified and I have to hit save again which kicks in the formatter again. So if I open a file and make a change there's potentially no way to get the file back to it's unchanged state without disabling the formatters since I always have to hit save even though I've undone all the changes.

[tyson-kubota]

• VSCode : Version 1.15.0 (1.15.0), 8b95971, 2017-08-09T20:02:33.839Z
• OS Version: macOS 10.12.6
• Extensions: none

A probably-related issue that snuck into a recent release:

Auto-formatting seems to now add an extra step in the undo history. So pasting into a (JavaScript) file will auto-indent the pasted portion, but pressing command-Z (or choosing Edit->Undo in the UI) only undoes the auto-indentation, and not the pasting itself.

The result is that it takes two invocations of undo to remove pasted content.

[bpasero]

#919

[tyson-kubota]

@bpasero I understand if the save/dirty-checking behavior is the same cause as #919 (I didn't realize this hole went so deep), but is my issue above related (#32582 (comment) — a recent regression with the latest release, as far as I can tell)?

[bpasero]

@tyson-kubota please keep one issue around one topic, if you see multiple issues, file them as individual issues.

[tyson-kubota]

Thanks. I now see that #32305 already expresses my issue.